AI cracks superbug problem in two days that took scientists years

Comments

• dojan@lemmy.world ⁨1⁩ ⁨day⁩ ago

  “I wrote an email to Google to say, ‘you have access to my computer, is that right?’”, he added.

  lmao right, because the support person he reached, if indeed he even spoke to a person at all, would know and divulge the sources they train on. Dude may think that all his research is private but they’re making use of these tech giant services. These tech giants have blatantly showed that they’re OK with piracy and copyright infringement to further their goals, why would spying on research institutions be any different?

  source

  • DarkCloud@lemmy.world ⁨1⁩ ⁨day⁩ ago

    Large Language companies weren’t even aware their data (which is so large they themselves have no idea what’s in it) had other languages.

    So the models suddenly knew how to speak other languages. The above story feels like those stories “Large Language Models are super intelligent! They’ve taught themselves French!” - no, mass surveillance and corporations being above the law taught them everything they know.

    source
  • A_A@lemmy.world ⁨1⁩ ⁨day⁩ ago

    (…) If you want to give it a run for its money, give it a novel problem that isn’t solved, and see what it comes up with.

    You mean like searcher have done …

    ... in here : ?

    bturtel.substack.com/p/human-all-too-human
    For AI to learn something fundamentally new - something it cannot be taught by humans - it requires exploration and ground-truth feedback.
    .
    www.lightningrod.ai
    We’re enabling self-play that learns directly from real world feedback.

    source
  • TrenchcoatFullofBats@belfry.rip ⁨1⁩ ⁨day⁩ ago

    “I wrote an email to Google Gryzzl to say, ‘you have access to my computer, is that right?’”, he added.

    Later that day

    source
• FlowVoid@lemmy.world ⁨1⁩ ⁨day⁩ ago

  Uh no, the AI didn’t crack any problem.

  The AI produced the same hypothesis that a scientist produced, one that the scientist considered his own original awesome idea.

  But the truth is that science is less about producing awesome ideas and more about proving them. And AI did nothing in this regard, except to remind scientists that their original awesome ideas are often not so original.

  There’s even a term scientists use when another scientist has the same idea but actually managed to do the work of proving it: “scooped”. It’s a very common occurrence.

  source
• homesweethomeMrL@lemmy.world ⁨1⁩ ⁨day⁩ ago

  Now if you’d all just empty your wallets into the AI bonfire. Thaaaat’s right.

  source
• A_A@lemmy.world ⁨1⁩ ⁨day⁩ ago

  it’s not word completion, its so far from it :

  (…) He told the BBC of his shock when he found what it had done, given his research was not published so could not have been found by the AI system in the public domain. (…)

  (…) “It’s not just that the top hypothesis they provide was the right one,” he said. "It’s that they provide another four, and all of them made sense. “And for one of them, we never thought about it, and we’re now working on that.” (…)

  source

  • DarkCloud@lemmy.world ⁨1⁩ ⁨day⁩ ago

    Assuming open AI ect only use data from the public domain is stupid. He has literally no idea what the AI has seen, but they’ve undoubtedly bought countless amounts of data that isn’t readily searchable by public engines.

    He sounds very ill informed on the matter of data collection and probably just had his info/data on a cloud service somewhere whose text was part of yhe trillions of terrabytes LLM have accessed.

    source

    • A_A@lemmy.world ⁨1⁩ ⁨day⁩ ago

      it seems you did not read my post in entirety.

      source
• SnotFlickerman@lemmy.blahaj.zone ⁨1⁩ ⁨day⁩ ago

  Google doesn’t need access to all his unpublished research if he’s ever mentioned anything about it online or in an email that went to a gmail address.

  Further, University of Cambridge runs on Microsoft Exchange and University of Glasgow uses Office365.

  Not to put to fine a point on it, but they don’t need access to your computer and this feels a little bit overhyped.

  Also just because it came to the same conclusion means about as much as it coming to the wrong conclusion, does it not? Since there is no actual “thinking” in these devices?

  source

  • Flisty@mstdn.social ⁨1⁩ ⁨day⁩ ago

    @SnotFlickerman @cm0002 unless he's done the research himself he won't know whether the results are viable - as he says, they've got to test the "new" one. So at best it gives you a bit of a head start on new avenues, at worst it completely wastes your time down a new rabbithole.

    source
• ryedaft@sh.itjust.works ⁨1⁩ ⁨day⁩ ago

  It’s so easy to ask a question in such a way that the statistically most likely answer is the one at the front of your mind.

  source
• jpreston2005@lemmy.world ⁨1⁩ ⁨day⁩ ago

  When AI decides to destroy the human virus, it now knows exactly how to create a bug capable of it. Probably more likely than pumping out a bunch of humanoid robots with guns, just create a bug, spread it around, and mess with our ability to communicate in time to stop the spread. BAM. Easy-peasy, humans are now down to a manageable 1 billion or so individuals.

  source
• ristoril_zip@lemmy.zip ⁨1⁩ ⁨day⁩ ago

  if this is machine learning and neural networks, I can believe it’s a good thing, maybe even meaningful for the potential of so called artificial intelligence.

  if this is an LLM that’s alleged to have popped this “virus tail” theory out of… what exactly…? I’m not buying it.

  source
• MNByChoice@midwest.social ⁨1⁩ ⁨day⁩ ago

  Great! We have a tested solution and scalled up th3 drug to treat the issue. And in 2 days! Great!

  Oh, that is not what we have?

  source